Semantic-Based Opinion Summarization

Proceedings of Recent Advances in Natural Language Processing, pages 619–628 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_070 619 Semantic-Based Opinion Summarization Marcio Lima In´acio and Thiago Alexandre Salgueiro Pardo Interinstitutional Center for Computational Linguistics (NILC) Institute of Mathematics and Computer Science (ICMC) University of S˜ao Paulo (USP) Avenida Trabalhador S˜ao-Carlense, 400, 13566-590 S˜ao Carlos, SP, Brazil marcio.lima.inacio@gmail.com, taspardo@icmc.usp.br Abstract The amount of information available online can be overwhelming for users to digest, spe- cially when dealing with other users’ com- ments when making a decision about buying a product or service. In this context, opinion summarization systems are of great value, ex- tracting important information from the texts and presenting them to the user in a more un- derstandable manner. It is also known that the usage of semantic representations can ben- efit the quality of the generated summaries. This paper aims at developing opinion sum- marization methods based on Abstract Mean- ing Representation of texts in the Brazilian Portuguese language. Four different methods have been investigated, alongside some litera- ture approaches. The results show that a Ma- chine Learning-based method produced sum- maries of higher quality, outperforming other literature techniques on manually constructed semantic graphs. We also show that using parsed graphs over manually annotated ones harmed the output. Finally, an analysis of how important different types of information are for the summarization process suggests that using Sentiment Analysis features did not improve summary quality. 1 Introduction With the advance of web technologies during the last decades, a numerous amount of textual data is produced constantly, especially User Generated Content within social media and e-commerce do- mains, which include opinions towards many en- tities, such as products, organizations, and others. Such quantity of information is virtually impos- sible for other users to assimilate by their own, which creates a need for automated content selec- tion and summarization methods, originating the Opinion Summarization research area, which inte- grates both the Text Summarization and Sentiment Analysis areas. (Kim et al., 2011; Liu, 2012). There are two main types of summaries: extracts, which are composed by copying excerpts from the original texts literally, and abstracts, consisting of new text constructions derived out of the informa- tion obtained from the input data. Some authors in the literature call attention to the need of incorpo- rating semantic knowledge into the summarization process (Mani, 2001; Li, 2015; Huang et al., 2020), particularly for producing abstracts, since humans usually create summaries by rewriting, paraphras- ing and, mainly, interpreting texts (See et al., 2017; Dohare et al., 2018). This kind of information can be explicitly formalized as semantic representa- tions, such as Predicate Argument Structure (PAS) (Khan et al., 2016, 2018), Abstract Meaning Rep- resentation (AMR) (Liu et al., 2015; Liao et al., 2018; Dohare et al., 2018) and others. There are also arguments promoting the usage of explicit semantic representations to deal with texts from the opinions genre – in which the sentiment expressed by the author is an important informa- tion – as this type of linguistic knowledge …

Content truncated to 500 words